PD126 - A STANDARDISED FRAMEWORK FOR ANALYSING REAL-WORLD DATA ON AUTOMATED MALNUTRITION SCREENING IN HOSPITALISED PATIENTS

PD126

A STANDARDISED FRAMEWORK FOR ANALYSING REAL-WORLD DATA ON AUTOMATED MALNUTRITION SCREENING IN HOSPITALISED PATIENTS

M. Gerwek^1,2,*, S. Walther¹, U. Holdgrün^1,2, A. Nierling³, L. Selig³, M. Stumvoll^3,4, H. Schlögl^3,4, T. Kirsten^1,2

¹Institute for Medical Informatics, Statistics and Epidemiology, Leipzig University, ²Dept. Medical Data Science, Medical Informatics Center, ³Department of Endocrinology, Nephrology and Rheumatology, University of Leipzig Medical Center, ⁴LeiCeM - Leipzig Center of Metabolism, Leipzig University, Leipzig, Germany

Rationale: Malnutrition remains frequently underdiagnosed among hospitalised patients, despite its clear association with adverse outcomes such as prolonged length of hospital stay and increased mortality. Structured screening tools, such as the Nutritional Risk Screening 2002 (NRS), have demonstrated clinical benefit when applied consistently, particularly in enabling early nutritional interventions.

Methods: This retrospective study analyses prevalence and outcomes of malnutrition risk using routine clinical data from a German university hospital (2013–2023). A standardised, R-based framework was developed, integrating automated data extraction, data quality checks, and harmonised statistical analyses. The framework enables local execution on site-specific data while generating aggregated, comparable outputs without sharing patient-level data. It includes case-level aggregation, validation of key variables, and standardised analyses, yielding cohort characteristics, prevalence estimates, subgroup stratifications, and associations with clinical outcomes.

Results: The findings of the study demonstrate that NRS-based screening identifies a high-risk subgroup characterised by consistently worse outcomes, including prolonged hospitalisation, elevated readmission rates, and increased mortality. Concurrently, substantial data heterogeneity, including missingness, inconsistent coding, and variability in NRS documentation require extensive preprocessing.

Conclusion: The project demonstrates the efficacy of a modular and standardised analysis framework in addressing the key challenges of routine data analysis while enabling scalable and privacy-preserving research. In addition, it emphasises the necessity for more robust frameworks that can systematically manage heterogeneous data and varying documentation practices to ensure reliable multi-centre analyses.

Disclosure of Interest: None declared